Upload folder using huggingface_hub (#2)

- 9e395a92d9afc9bcc58ef1ed9092513b24b5439432f4a911fc9f13a9d8a7b638 (ca7da98bb607b824f6351328b050e55cbb2347f1)
- b17547f69c7d0703f85eabd6eb6d4067778dccc9b9c205af8394959812f898f9 (fb63c06d7fa1374fe8d2437dea63faf5ae8338f5)
- ead3442caccc6db2a697fe5b8d44c19c21fdf41ea99d30a7bf403de6f933a2e3 (2f985b76d5074f58d4ecda3e6134ca4df0e64264)

README.md

@@ -34,7 +34,7 @@ tags:
 
 ## Results
 
-Detailed efficiency metrics coming soon!
+![image info](./plots.png)
 
 **Frequently Asked Questions**
 - ***How does the compression work?*** The model is compressed with llm-int8.
@@ -61,7 +61,7 @@ You can run the smashed model with these steps:
    from transformers import AutoModelForCausalLM, AutoTokenizer
 
     model = AutoModelForCausalLM.from_pretrained("PrunaAI/migtissera-SynthIA-7B-v1.3-bnb-8bit-smashed",
-                                                 trust_remote_code=True)
+                                                 trust_remote_code=True, device_map='auto')
     tokenizer = AutoTokenizer.from_pretrained("migtissera/SynthIA-7B-v1.3")
     
     input_ids = tokenizer("What is the color of prunes?,", return_tensors='pt').to(model.device)["input_ids"]

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "/tmp/tmpv6r32k1i",
+  "_name_or_path": "/tmp/tmp9iof9sod",
   "architectures": [
     "MistralForCausalLM"
   ],
@@ -18,7 +18,7 @@
   "quantization_config": {
     "bnb_4bit_compute_dtype": "bfloat16",
     "bnb_4bit_quant_type": "fp4",
-    "bnb_4bit_use_double_quant": true,
+    "bnb_4bit_use_double_quant": false,
     "llm_int8_enable_fp32_cpu_offload": false,
     "llm_int8_has_fp16_weight": false,
     "llm_int8_skip_modules": [

smash_config.json

@@ -8,7 +8,7 @@
         "compilers": "None",
         "task": "text_text_generation",
         "device": "cuda",
-        "cache_dir": "/ceph/hdd/staff/charpent/.cache/modelseisy_ix4",
+        "cache_dir": "/ceph/hdd/staff/charpent/.cache/models5mrssbi8",
         "batch_size": 1,
         "model_name": "migtissera/SynthIA-7B-v1.3",
         "pruning_ratio": 0.0,